基于核空间距离测度的特征选择

提出核空间距离测度这一可分性判据。在核空间中计算两类样本点之间的距离，并以距离的大小评价子集的分类性能。使用顺序前进法作为搜索算法，在人造和真实的数据集上进行测试，文中的核空间距离测度可分性判据明显优于传统非核的可分性判据，优于或接近于Wang提出的核散布矩阵测度，并在运行时间上快一个数量级。将文中方法应用于胰腺内镜超声图像分类，取得较好分类结果。

Feature Selection Algorithm Based on Kernel Distance Measure

The kernel distance measure is proposed as a new type of class separability. The distance of samples between two classes is measured in the kernel space and used to evaluate the separability of subsets. Using the sequential forward selection algorithm as the search strategy, tests are carried out on both synthetic and real datasets. Experimental results demonstrate that the proposed method outperforms the traditional non-kernel class separability. Moreover, the proposed method is superior or close to the kernel scatter matrix measures proposed by Wang and its running time is an order of magnitude faster. When applied to the pancreatic EUS image classification, the proposed method receives a good result.